`
`
`
`Downloaded on 2017-01-21 to IP 128.113.122.202 address. Redistribution subject to ECS terms of use (see ) unless CC License in place (see abstract). ecsdl.org/site/terms_use
`
`
`
`
`
`
`
`Page 1 of 7
`
` IP Bridge Exhibit 2032
` TSMC v. IP Bridge
` IPR2016-01246
`
`
`
`Chemical-Mechanical Polishing for Fabricating Patterned W Metal Features as Chip Interconnects F. B. Kaufman, 1 D. B. Thompson, 2 R. E. Broadie, 2 M. A. Jaso, *'l W. L. Guthrie, 2 D. J. Pearson, ~ and M. B. Small 1 IBM Research Division, Thomas J. Watson Research Center, Yorktown Heights, New York 105981 and IBM General Technology Division, Hopewell Junction, New York 109532 ABSTRACT Interconnect features of W metal, recessed in an SiO2 dielectric, can be formed using a novel chemical-mechanical pol- ish process. Mechanical action, to continually disrupt a surface passivating film on W, and chemical action, to remove W, appear to be requirements for workability of the process. A trial process chemistry using a ferrieyanide etchant is de- scribed. Removal of the W is discussed in terms of competition between an etching reaction which dissolves W and a pas- sivation reaction to reform WOa on the surface of the W. This novel processing technology is compared with earlier meth- ods of fabricating metal interconnect structures. It is now generally recognized that planarization of mul- tilevel metal interconnect structures for chips offers signif- icant process advantages such as the elimination of step- coverage concerns and improvement of lithographic reso- lution by minimizing depth of field variations (1-7). Two al- ternative process approaches to achieving planarization can be considered (Fig. 1). One approach relies on dielec- tric planarization. In this case patterned metal intercon- nects are conformally covered with an insulator film. This is followed by a planarizing operation to eliminate the to- pography in the dielectric. Processes that have been used for smoothing include etch-back (8) of the dielectric, depo- sition of a planarizing polymer film (9), or chemical- mechanical polishing (10). Alternatively, recessed metal schemes (1-5) can be used where the reverse of the metal pattern desired is etched into a planarized film of dielec- tric, metal is conformally deposited, and then subse- quently removed, in a separate step, from the higher blanket areas on top of the planarized dielectric to leave the required metal pattern recessed in the insulator. An improved low-pressure chemical vapor deposition (LPCVD) process (11, 12) for tungsten, coupled with its high electromigration resistance and dry etch compatibil- ity, makes it an attractive candidate for use in a recessed metal interconnection sequence (3, 5). However, current reactive ion etching (RIE) processes for tungsten which have been used to remove the blanket metal suffer from selectivity concerns due to resist thickness variations over the metal (3). Wet etching techniques are unacceptable due to their inability to preferentially remove topographic fea- tures. Chemical-mechanical polishing (1, 2) could be used to define the tungsten features if a process chemistry appro- priate for wafer fabrication were available. Chemical- mechanical polishing of metals has been demonstrated for stainless steels and nickel-based alloys (13), and for copper (2, 14). The mechanism we propose for the process (Fig. 2) requires the action of a metal etchant and a metal passi- vating agent with an abrasive agent. This combination could result in the high spots continually having the passi- vated film (anisotropically) etched away, while the low spots are protected. For the majority of metals, the oxide may be used as a simple passivant. During the process, the protective film is removed by the mechanical action of the abrasive slurry. This is followed by a rapid reformation of the protective film. Continuous cycles of formation, re- moval, and reformation of the passivating layer continue until the desired final thickness of metal is achieved. Consideration of the proposed mechanism and Fig. 2 suggests that the minimal requirements for the proposed process chemistry for W removal are: (i) materials selectiv- ity, a significantly faster removal rate for the W than either the dielectric surface, which forms the structure, or a sacri- ficial etch stop; (ii) topographic selectivity, a metal re- moval process which selectively removes metal from the * Electrochemical Society Active Member. "high" spots while leaving it protected in the low spots; (iii) the overall process should be noncorrosive, (iv) the process should leave the wafers clean enough to be com- patible with further semiconductor processing in a clean room. This paper will describe a process chemistry that has been successfully used in the chemical-mechanical pol- ishing of tungsten to form chip interconnect structures using a recessed metal process sequence. Given the nature of the polish mechanism described, we suggest that the particular etchant-passivator combination discussed here is not the only chemical system that could be used for suc- cessful fabrication of interconnect structures. Preliminary descriptions of the application of this technology to form a fully planarized interconnect structure with 1.2 ~tm con- tacts, as applied to a 64 Kb complimentary metal oxide semiconductor static random access memory (CMOS SRAM), and, using x-ray lithography to achieve 0.5 #m on all interconnect levels, have previously appeared (6, 7). Experimental Materials and preparation of slurry.--All chemicals were used without further purification. Typical as- received purity levels were at least 98%. Water used in the preparation of the slurry had resistivities greater than 18 M~ and was filtered through a Millipore filter system, type RO to remove ionic and organic contaminants and particles less than 1 ~m in size. DIELECTRIC RECESSED PLANARIZATION (DP) METAL (RM) DP-I. PATTERNED METAL RM-I. PATTERNED DIELECTRIC --,~_f-m, 1 DP-2. CONFORMAL RM-2. CONFORMAL DIELECTRIC METAL DEPOSITION DEPOSITION DP-& PLANARIZATION OF RM-5. PLANARIZATION OF DIELECTRIC METAL Fi 9. 1. Comparison of dielectric planarization (DP) and recessed metal (RM) approaches to the fabrication of chip interconnect struc- tures. 3460 J. Electrochem. Soc., Vol. 138, No. 11, November 1991 (cid:14)9 The ELectrochemical Society, Inc.
`
`
`
`Downloaded on 2017-01-21 to IP 128.113.122.202 address. Redistribution subject to ECS terms of use (see ) unless CC License in place (see abstract). ecsdl.org/site/terms_use
`
`
`
`
`
`
`
`Page 2 of 7
`
`
`
`J. Etectrochem. Sac., Vol. 138, No. 11, November 1991 (cid:14)9 The Electrochemical Society, Inc. 3461 MECHANICAL ACTION (2)~ WET ETCH OF UNPROTECTED METAL BY CHEMICAL ACTION; PASSIVATING FILM REFORMS (3) S~ ' ]i!pAO / (3) | PLANARIZATION BY REPETITIVE CYCLES OF (I) AND (2) METAL//~/ INSULATOR ~,/////~.~ Fig. 2. Proposed mechanism of planadzation of a patterned metal feature by chemical-mechanical polishing (see text far details). All slurries were prepared by adding the chemical con- stituents to a rapidly stirred solution of water. Addition of the abrasive to the solution typically resulted in the solu- tion becoming optically opaque. To insure homogeneity of the polishing slurry, the resulting suspension was stirred during use. Although there was evidence for precipitate formation in the slurry after several hours, polish results were not found to change after storage for times up to 8 h. All slurries were delivered to the polish table using a peri- staltic pump. Slurry particle size analysis.--A Coulter Model LS 130 laser diffraction particle size analyzer was used to measure the particle size of the free abrasive in the polish slurry. Size calibration was checked using appropriate standards in the 0.1-100 ~m particle size regime. Polish slurry parti- cle sizes were determined on flowing samples (recir- culation rates of 0.5 liter/rain) using dilution factors (in water) in the 0.1 to 0.5 range. Chemical-mechanical polish tools, fixtures, and operat- ing parameters.--In general, we expect the same polish tools and fixturing traditionally used in the chemicai- mechanical polishing of blanket Si wafers, to be applicable to the metal polishing process described in this work. In Fig. 3, is shown a schematic of the key components of the polish technique. The wafer (i) is placed in a holder (2) with the wafer surface in direct contact with a pad-covered table (4). During the polish experiment, the abrasive slurry (3) flows onto the surface of the pad and the rotation speed of the table and of the holder can be independently varied. Pressure at the wafer-slurry-pad interface is controlled via an overarm mechanism which allows pressures in the 1 to 5 kg/cm 2 range to be applied to the wafer holder. Most of the data presented here was obtained on Strasbaugh Model 6CA and 6DQ polishers but polish tools from other manufacturers are expected to be roughly equivalent in overall performance. Several different types of polish pad materials and wafer template assemblies obtained from Rodel were found to provide acceptable and reproducible results. Fixturing provided by other suppliers is expected to give similar results. Si and W blanket and patterned substrates.--All pol- ishing experiments were done on 125 mm Si wafers using an SiO2 insulator layer. Following the deposition of a sput- Fig. 3. Schematic of chemical mechanical-polishing technique, top and side views. For clarity, overarm mechanism connected to wafer holder has been omitted. Wafer (1) is held in holder (2) using commer- cially available template; slurry (3) flows between wafer surface and pad (5) covered table (4). See text for further details. tered Ti/TiN adhesion layer, W was deposited using an LPCVD process (11, 12) in a batch CVD reactor (Genus 8402). Patterned wafers (6) for demonstration of the re- cessed W chemical-mechanical polishing process were prepared either by the Yorktown or East Fishki]l Si Facil- ity. These wafers had W deposited into RIE patterned fea- tures of a planarized SiO2, with the planarization achieved using a previously described (10) chemical-mechanical polish process to smooth topographic variations in the SiQ insulator. Metallization consisted of RIE patterned Ti/Al(2.5%Cu)/Si deposited by dc magnetron sputtering. Argon sputter-cleaning was used prior to the metal deposi- tion to insure adequate adhesion to the W and to the insu- lator surfaces. W film thickness;determination.--Average metal thick- ness and standard deviation was determined using either a manuaI four-point sheet resistivity probe apparatus or an automated four-point Prometrix Model 20 measurement tool. Thicknesses of the films were calculated from the ex- perimentally measured sheet resistivity using the previ- ously determined value for the layer resistivity (12) of the LPCVD W. Determination of wet etch rates.--Removal rates of W were determined by comparing initial thicknesses (as de- termined above) of the W films vs. the thickness remaining after 1-24 h of contact with the liquid of interest. All results were determined for static solutions at 20~ Results and Discussion In the absence of complexation reagents or substances which form insoluble salts, W becomes passivated at acidic pH values less than 4 (15) due to the formation of WO3. However, in the presence of an oxidant such as K3Fe(CN)6 and weak organic base complexing agents, the range of pH in which W gets passivated is extended to 6.5 (16). Solutions were formulated containing the weak oxi- dant KaFe(CN)6, and it was observed that by adjusting the pH, both low static wet etch rates and chemical-mechani- cal polish removal of W films could be simultaneously achieved, see Table I. We observed that when the chemical formulations were combined with silica or alumina abrasive particles, W films could be chemical-mechanically polished at removal rates as high as 400 nm/min. For the formulations listed in Table I with pH values less than 6.5, the W removal rates obtained were found to be functions of the "mechanical" components of the process, viz., rates increased with ap- plied pressure, quill/table rotation speeds, and increased
`
`
`
`Downloaded on 2017-01-21 to IP 128.113.122.202 address. Redistribution subject to ECS terms of use (see ) unless CC License in place (see abstract). ecsdl.org/site/terms_use
`
`
`
`
`
`
`
`Page 3 of 7
`
`
`
`3462 J. Electrochem. Soc., Vol. 138, No. 11, November 1991 (cid:14)9 The Electrochemical Society, Inc. Table 1. Formulation Wet etch rate CMP observed (Note 1) pH (nm/sec) (Note 2) K-F 5.0 8 + K-F-En 6.5 6 + F-En 10.5 43 - F = K~Fe(CN)~; K = KH~PO4; En = Ethylenediamine(cid:12)9 1. Formulations typically contain 0.1-5 weight percent of chem- icals. 2. Chemical-mechanical polishing (CMP) and pH determined on formulations containing abrasive. (See text for details.) loading of abrasive. In the absence of applied pressure or free abrasive, the W removal rates were found to be mini- mal. This implies that areas of recessed metal would only be slowly removed. However, for the chemical formulation with pH 10.5, significant removal rates were found in the absence of added slurry or applied pressure suggesting, in this case, a removal process which is hydrodynamically as- sisted wet etching rather than chemical-mechanical pol- ishing. The optimum particle size and particle size distribution for the abrasive additive has not been determined. How- ever, we have observed that commercially available colloi- dal silica or dispersed alumina in the 0.2-2.0 ~m size range (see Experimental section) yield W surfaces after polishing that are highly specular and which appear scratch-free to the human eye. In Fig. 4 is shown particle size data for an alumina dispersion. We find that for any of the process chemistries in Table I, the measured particle size and dis- tribution stays relatively constant. Although the alumina is specified as a deagglomerated solid with 0.05 ~m aver- age particle size, the light scattering measurements sug- gest that in the flowing, liquid state, the average particle size is considerably larger. Whether alumina aggregates form in solution and subsequently break apart during the polishing operation is not known at present. We have ob- served that addition of a coagulation agent to the slurry re- sults in immediate settling of the normally dispersed fine particulates. Particle size analysis performed on these co- agulated samples show significantly larger measured par- ticle size, in addition to a distinct increase in surface scratching of the metal films polished under these condi- tions. This suggests a potentially important relationship between average particle size at the wafer surface and the incidence of polish-induced damage at that surface. Further distinction between removal of a blanket W film under chemical-mechanical polish conditions vs. wet etch- ing (when process chemistry pH is above the passivation limit) can be made by observing the thickness decrease over the surface of the wafer as a function of polish process time. In Fig. 5 are shown two traces where W thickness is plotted over multiple points on the wafer surface before (a) and after (b) 3 min of polish processing in the chemical- mechanical polish mode. Two effects are observed which reflect the novel aspects of the removal process. It is seen that during the polishing, in addition to a thinning of the metal film as a function of time, the thickness uniformity of the metal can be modified. In the specific example shown in the figure, after 3 rain of polishing at 130 nm/min the original film, which was lower in the center, now has a 5 ~ i , ,i , i i , ,i i i , i i i I I I , O.Z O.4 1.0 2 4 6 I0 20 40 IO02OOzlOO IOO0 PARTICLE DIAMETER (/.r Fig. 4. Particle diameter in microns vs. volume percent population for an alumina slurry in the F-K-En chemistry as determined by light scat- tering. (a) .. :-/y - =.. (b) ........ 2 \X \,. ", '..t"-,\ i .~ \',,\_X'~_\',~ ',.. "-. ~ " \ ~ '~ '\ slllt " "I t 3 J I- / / -;' I , ! ,, I I -~__..1-...Jll/i / / i t i i ~ " J~'- ,/l I I I ,." # / , [ / ~--~-3LI ),; ) ) / ] , ' / / / I \_. _ _ , / / /, /, jr ,, ./ / ~ ; '~, ~"~.~~ _ I / ff /," ,, ,' ' /" / .... -.7-'~ __/ ..... / / Fig. 5. Sheet resistivity contour maps of blanket W films, where lines connect measured points of constant resistivity. Bold solid lines give mean resistivity with dotted lines showing areas which have higher (+) or lower (-) sheet resistivity than the mean. (a) Initial film with mean, calculated, thickness of 500 nm; each contour interval (dotted line) has successively 4% higher or lower sheet resistivity than the mean. (b) after 3 min polishing, mean thickness 112 nm with 8% deviation per contour line. wedge shape with the maximum to minimum thickness variation observed from one wafer edge to another(cid:12)9 In gen- eral, we have found that many different kinds of final global planarity can be achieved, dependent on the initial thickness uniformity of the metal film and the settings of the polish tool. Experiments done using identical polish tool settings but with the F-En chemistry at pH 10, show an isotropic removal of W, with no change in the global planarity of the film, typical of a wet etching process. In addition to the sensitivity of the removal process to choice of mechanical parameters during polishing, as mentioned above, we find that the polish rates also depend on the concentration of chemicals listed in Table I and on the temperature of the slurry during processing. This sug- gests, consistent with the proposed mechanism, that the polish process is driven both by mechanical and by chemi- cal effects. We typically observe that polish rates are pro- portional to the concentration of chemical constituents listed in Table I. In addition, we find that at any given con- centration of chemicals, polish rates increase as a function of increasing temperature of the slurry. By controlling the temperature of the recirculating water which flows through the polish table the temperature of the slurry con-
`
`
`
`Downloaded on 2017-01-21 to IP 128.113.122.202 address. Redistribution subject to ECS terms of use (see ) unless CC License in place (see abstract). ecsdl.org/site/terms_use
`
`
`
`
`
`
`
`Page 4 of 7
`
`
`
`J. Electrochem. Soc., Vol. 138, No. 11, November 1991 (cid:14)9 The Electrochemical Society, Inc. 3463 This process can readily be applied to form W vias in an SiO~ insulator layer (6, 7). A typical process sequence in- volves deposition of the insulator layer followed by litho- graphic patterning of the insulator, sequential deposition of both an adhesion layer and blanket W, followed by pol- ishing to remove the surface W. Figure 6 is a scanning elec- tron micrograph top down view of W patterns that result after processing. We observe that the patterned metal is well defined with no evidence of metal smearing into the adjacent insulator. In addition, the surface of the W is highly specular without any obvious chemical attack or corrosion. Cross-sectional scanning electron micrograph (see Fig. 7) shows clear evidence for the planarity of the pro- cess and indicates that the W seams, unavoidable with a conformal CVD deposition process, have not been de- graded by process induced wet etching. In general, we find that the degree of local planarity (metal vs. insulator height) and the final thickness and uniformity of the insu- lator film is dependent on the removal rate ratio of metal- to-SiO~ insulator. However, we have observed that subtle changes made to the process chemistry can lead to a degradation in quality of the recessed metal surface. This is an effect only ob- served on patterned wafers since blanket, polished films are always smooth. On blanket W wafers the K-F chemical system, with or without the addition of ethylenediamine, shows very similar blanket polish rates, and low static wet etch rates suggesting efficient surface passivation. Pat- terned wafers polished with the K-F-En chemistry show high-quality cross-sectional SEM features similar to those in Fig. 7. However, patterned wafers polished under ex- actly the same conditions and using the K-F chemistry in the absence of ethylenediamine show distinct signs of sur- face attack and loss of planarity relative to the adjacent in- sulator surface. This observation suggests that in the pres- ence of ethylenediamine, and under the dynamic conditions of polishing, surface passivation of W is more efficient than in the absence of the weak amine base. The mechanism proposed for W passivation in the presence of ferricyanide involves (16), see Eq. [1] W + 6Fe(CN~ 3 + 4H20 ~ WO~ 2 + 6Fe(CN)~ 4 + 8H + [1] the generation of protons at the surface via a W/W +6 redox reaction. The local concentration of protons at the metal interface can be expected to be influenced by the presence of buffering agents and weak base. We, therefore, suggest that static wet etch rates and bulk solution pH values alone do not entirely reflect the dynamic chemical environment which occurs at the metal-solution interface during metal polishing. Fully planarized, two level interconnect structures, with W studs forming the contacts to the underlying Si device areas and between the wiring levels i.e. (W studl M1-W stud2-M2), have been recently fabricated (6, 7). The W stud levels were formed in a chemical-mechanical polish pro- cess using the K-F-En process chemistry. Overall yields were high and contact resistance values were not affected by the chemical-mechanical polish processing. The data show, see Fig. 8, the contact resistance scales with feature Fig. 6. SEM views of a patterned W in Si02 dielectric surface follow- ing chemical-mechanical polishing. tacting the wafer surface can be controlled. Given the hy- drodynamic and chemical complexity of the polishing pro- cess, we suggest that modeling of the concentration and thermal effects observed will have to take into account a number of competing processes. These include diffusion to and depletion of the chemical constituents at the wafer interface, the kinetics of passivating film formation, and the dynamic rate of isotropic wet etching (see below) prior to the formation of the passivating film. Fig. 7. Cross-sectional SEM view of W studs in SiOz insulator.
`
`
`
`Downloaded on 2017-01-21 to IP 128.113.122.202 address. Redistribution subject to ECS terms of use (see ) unless CC License in place (see abstract). ecsdl.org/site/terms_use
`
`
`
`
`
`
`
`Page 5 of 7
`
`
`
`3464 J. Electrochem. Soc., Vol. 138, No. 11, November 1991 (cid:14)9 The Electrochemical Society, Inc. I0 4. MI/WIx-" ,.... Z 1 ",. ~ 1.0 1,~~,,li:,. ................ I,I 0: I-- "1 "'., Z~ 5.exg.Oum O.I 0 ~ "~ i i l I IlllL ~ ] I I Illl I I I I I I I II I I0 I00 W FEATURE AREA (F.m 2) Fig. 8. Contact resistance of W contacts recessed in an SiO2 dielec- tric for feature sizes from 0.5 to 5.0 ~m 2. Comparison made for W con- tacts from Ti/AI(2.5%Cu)/Si Mi to silicided n +, p+, and to n + polysili- con chip regions. area for 0.5 to 5.0 ~m 2 contacts. This indicates either the absence of significant polish-induced contamination at the interface or that the surface contamination produced does not survive the sputter-clean step performed prior to met- all deposition (see the Experimental section). M1-M2 via resistances were appropriately lower due to the absence of the silicide. An improved W chemical-mechanical pol- ishing process was used to fabricate a 4 Mbit DRAM (17) structure using a CVD-tungsten bit line and contacts in a dual damascene (18) process in manufacturing. A consequence of the low contact resistance values ob- served for the W interfaces and the proposed mechanism of chemical-mechanical polishing is the presence of a chemically passivating layer on the surface of the polished tungsten. XPS and Auger analysis of polished CVD blanket W samples confirms (19) the presence of a thin layer of tungsten oxide. It is estimated that the oxide con- sists of 5-6 A of WO~ over 3-10 A of some other lower oxi- dize that forms a transitional region between the WO3 and the bulk W. Furthermore, there was no evidence to indi- cate the presence of any significant post polish residue or other impurities. We propose that the chemical-mechanical polishing of W proceeds initially (see Fig. 2) by the abrasive removal of this protective thin oxide film. Once the unpassivated sur- face of the W film is exposed by the mechanical process, the W is removed by the chemical activity of the ferricya- nide/hydrogen-pho sphate/ethylenediamine combination. In this interpretation, ferricyanide acts as an electron sink to oxide and solubilize the W as a WO~ 2 specie (see reaction [1]). Competing with this etching reaction is reaction [2], to reform a new layer of the passivating oxide W + 6Fe(CN)g 3 + 3H20 ---> WO3 + 6Fe(CN)~ ~ + 8H § [2] Direct evidence for the intermediacy of the Fe(III) moi- ety and its consumption during polishing is obtained by observing the color changes which occur during the polish process. For example, when polishing a blanket W wafer we observe that the initial yellow color of the ferricyanide species is replaced, while the W is being consumed, by a much more colorless, opaque, solution. Following com- plete removal of the W, the yellow color seen initially reap- pears. We interpret these color changes to mean that while W is being actively dissolved (reaction [1]) or consumed to form a new passivating layer (reaction [2]), the colored Fe(III) species rapidly reacts to form the colorless Fe(II) re- duction product. Once all the W is removed in the polish process, the concentration of the Fe(III) species again in- creases and the original color returns. In this interpreta- tion, the ferricyanide is responsible for the oxidation of the W, while the hydrogen phosphate buffer-ethylenediamine base combination acts, at the W-solution interface, to con- trol the local pH. Thus once the passivating film is dis- rupted by a mechanical event, at a given dynamic concen- tration of ferricyanide oxidant, the competition between the etching and passivation reactions is determined by the interfaeial concentration of buffer and weak base. High quality, patterned surfaces of W can only be achieved when there is the appropriate dynamic balance between the W etch and passivation effects. Conclusions A chemical-mechanical polish process for the removal and planarization of W films is described. The process in- volves polishing with a mixture of ferricyanide-phosphate in the presence of free abrasive particles. Mechanical ac- tion to continually remove a passivating fil